Social networks of lexical innovation

Load code

Load data

single lemma

metrics for all lemmas

Filter data

Uses in first subset

df_comp %>%
  filter(
    SUBSETTING == 'time',
    SUBSET == 'one'
    ) %>%
  select(LEMMA, USES_TOT, USES) %>%
  arrange(USES)

Case studies

Check tweets

tweets %>%
  select(tweet, date) %>%
  # slice(., sample(1:n())) #random selection
  arrange(date)

Usage intensity

uses <- get_uses(tweets)
uses_tot <- get_uses_tot(uses)
age = get_age(uses)
coef_var <- get_coef_var(uses)
mean_date <- get_mean_date(uses)
max_date <- get_max_date(uses)
uses_month <- conv_uses_month(uses)
uses_plt <- plt_uses(uses_month, lemma, mean_date, max_date)
ggplotly(uses_plt)

Degree centralization

Diachronic

plt <- df_comp %>%
  select(LEMMA, SUBSETTING, SUBSET, CENT_DEGREE) %>%
  filter(
    SUBSETTING == 'time',
    SUBSET != 'full',
    LEMMA %in% cases
  ) %>%
  mutate(SUBSET = factor(SUBSET, levels=c('one', 'two', 'three', 'four'))) %>%
  ggplot(., aes(x=SUBSET, y=CENT_DEGREE)) + # group=1
    geom_point(aes(group=LEMMA, color=LEMMA, shape=LEMMA)) +
    geom_line(aes(group=LEMMA, color=LEMMA, linetype=LEMMA)) +
    guides(group=FALSE) +
    ggtitle('Diffusion over time: changes in degree centralization') +
    scale_y_continuous('degree centrality (log)', trans='log') +
    scale_x_discrete('subset')

plt

# ggsave('out/cent_deg_diac_Anglo-Saxon.pdf', width=6, height=4)

Overall

df_comp %>%
  filter(
    LEMMA %in% cases,
    SUBSET == 'full'
    ) %>%
  select(
    LEMMA, 
    EDGES,
    CENT_DEGREE, 
    CENT_EV
    ) %>%
  arrange(desc(CENT_EV))

Comparative analyses

Degree centrality

Overall

List

df_comp %>%
  select(LEMMA, SUBSET, USES, CENT_DEGREE, CENT_EV) %>%
  filter(
    SUBSET == 'full',
    USES >= 2
    ) %>%
  arrange(
    (CENT_DEGREE)
    # desc(CENT_EV)
  )

Plot

plt <- df_comp %>%
  select(LEMMA, SUBSET, USES, CENT_DEGREE) %>%
  filter(SUBSET == 'full') %>%
  arrange((CENT_DEGREE)) %>%
  ggplot(., aes(x=CENT_DEGREE, y=reorder(LEMMA, CENT_DEGREE))) +
    geom_point() +
    scale_y_discrete('lemmas') +
    scale_x_continuous(
      'degree centralization (log)',
      trans='log'
      )

plt

# ggsave('out/cent_sync_all.pdf', width=6, height=4)

Over time

Across all lemmas

df_comp %>%
  filter(
    SUBSET %in% (c(
      'one', 
      'two',
      'three',
      'four'
      )),
    USES_TOT >= 10000
    ) %>%
  group_by(SUBSET) %>%
  summarize(CENT_AVG = mean(CENT_DEGREE)) %>%
  mutate(SUBSET = factor(SUBSET, levels=c(
    'one', 
    'two',
    'three',
    'four'
    ))) %>%
  ggplot(., aes(x=SUBSET, y=CENT_AVG, group=1)) +
    geom_line() +
    geom_point()

Biggest changes

df_comp %>%
  select(LEMMA, SUBSET, CENT_DEGREE, EDGES, USES_TOT) %>%
  filter(
    SUBSET %in% c(
      'one', 
      'four'
      ),
    USES_TOT >= 10000
    ) %>%
  dplyr::group_by(LEMMA) %>%
  dplyr::mutate(CENT_DIFF = CENT_DEGREE - lag(CENT_DEGREE)) %>%
  drop_na() %>%
  select(-SUBSET) %>%
  rename(
    CENT_LAST = CENT_DEGREE,
    EDGES_LAST = EDGES
    ) %>%
  arrange((CENT_DIFF))

Usage intensity

df_comp %>%
  filter(SUBSET == 'full') %>%
  arrange(desc(USES))

Usage intensity vs. network characteristics

Uses vs. degree centralization

Plot

plt <- df_comp %>%
  filter(
    SUBSET == 'four'
    # USES %in% (10000:2000000),
    # USES %in% (150000:500000),
    # !LEMMA %in% c('slut shaming', 'dashcam', 'shareable', 'cuckold', 'deep learning', 'hyperlocal')
    ) %>%
  select(LEMMA, CENT_DEGREE, USES, EDGES) %>%
  ggplot(., aes(x=CENT_DEGREE, y=USES)) +
    geom_text(aes(label=LEMMA), hjust=-0.1, vjust=-0.1) + 
    geom_point() +
    scale_y_continuous(
      'usage frequency (log)', 
      trans='log'
      ) +
    scale_x_continuous(
      'degree centralization',
      # trans='log'
      ) +
    ggtitle(
      # 'Usage frequency vs. degree centralization: full sample'
      'Usage frequency vs. degree centralization: zooming on case study lexemes'
      )
    # geom_smooth(method=lm)

ggplotly(plt)

# ggsave('~/Desktop/freq-vs-net.pdf', width=6, height=4)
    

Biggest discrepancies

df_comp %>%
  filter(
    SUBSETTING == 'time',
    SUBSET == 'four'
    ) %>%
  select(LEMMA, USES_TOT, SUBSET, USES, CENT_DEGREE) %>%
  mutate(DISC = USES / CENT_DEGREE) %>%
  arrange(DISC)

Correlation

df_corr_full <- df_comp %>%
  filter(
    # SUBSET != 'full'
    # EDGES >= 100
    ) %>%
  select(-c(LEMMA, SUBSET, START, END, SKIP, STAMP, NROWS))
  
cor.test(df_corr_full$USES, df_corr_full$CENT_DEGREE)

    Pearson's product-moment correlation

data:  df_corr_full$USES and df_corr_full$CENT_DEGREE
t = -2.8574, df = 549, p-value = 0.004434
alternative hypothesis: true correlation is not equal to 0
95 percent confidence interval:
 -0.20253487 -0.03790525
sample estimates:
       cor 
-0.1210524 

Degree centrality vs. communities

Correlation

df_comp %>%
  filter(SUBSET == 'last') %>%
  select(CENT_DEGREE, COMMUNITIES) %>%
  mutate(COMMUNITIES = as.numeric(COMMUNITIES)) %>%
  correlate()

Plot

df_comp %>%
  filter(SUBSET == 'last') %>%
  select(LEMMA, CENT_DEGREE, COMMUNITIES) %>%
  ggplot(., aes(x=CENT_DEGREE, y=as.numeric(COMMUNITIES))) +
    geom_text(aes(label=LEMMA)) +
    scale_x_continuous(trans='log')

Uses vs. users

Plot

plt <- df_comp %>%
  filter(SUBSET == 'full') %>%
  select(LEMMA, USES, USERS) %>%
  ggplot(., aes(x=USERS, y=USES)) +
    geom_text(aes(label=LEMMA)) +
    scale_x_continuous(trans='log') +
    scale_y_continuous(trans='log') +
    geom_smooth(method=lm)
ggplotly(plt)

Correlation

df_comp %>%
  filter(SUBSET == 'full') %>%
  select(USES, USERS) %>%
  correlate()

Coefficient of variation

df_comp %>%
  filter(
    SUBSET == 'full',
    USES >= 1000
    ) %>%
  select(LEMMA, USES, COEF_VAR) %>%
  arrange(desc(COEF_VAR))

Processing status

Lemma list

df_comp %>%
  select(LEMMA, SUBSET, STAMP) %>%
  filter(SUBSET == 'last') %>%
  # mutate(STAMP = as_datetime(STAMP)) %>%
  arrange(desc(STAMP))

Dataset statistics

df_comp %>%
  filter(SUBSET == 'full') %>%
  select(LEMMA, SUBSET, USES, USERS) %>%
  dplyr::summarise(
    USES_TOT = sum(USES),
    USERS_TOT = sum(USERS)
    )

COEF_VAR vs. CENT

df_comp %>%
  filter(SUBSET == 'full') %>%
  select(LEMMA, COEF_VAR, CENT_DEGREE) %>%
  ggplot(., aes(y=COEF_VAR, x=CENT_DEGREE)) +
    geom_text(aes(label=LEMMA)) +
    scale_y_continuous(trans='log')

Correlations: EDA

library(Hmisc)

df_corr <- df_comp %>%
  # filter(SUBSET == 'last') %>%
  select(-c(LEMMA, SUBSET, NET_WINDOW_DATES, SKIP, STAMP, NROWS))
  # select(-c(USERS, AGE)) %>%
  # mutate(FOCUS = USES) %>%
  # focus(FOCUS) %>%
  # ggplot(., aes(reorder(rowname, FOCUS), FOCUS)) +
    # geom_col() +
    # coord_flip()
  # rearrange() %>%
  # shave() %>%
  # rplot()
  # network_plot(min_cor=.5) %>%